One question I have is- can we expect GPTQ quantization down to 4 bits (or lower) for these models, the same as for llama (which is partly why llama was so amaz

One question I have is- can we expect GPTQ quantization down to 4 bits (or lower) for these models, the same as for llama (which is partly why llama was so amazing?) I'm sure the answer is "yes", but I wonder if there's anything that can be done that could make these models more "friendly" to quantization like this from the very beginning. It's absolutely incredible that you can cut the size of a model in half... and then cut the size of the model in half *again*... and it fits on your consumer GPU with only a small cut in accuracy.